前言

mpstat相比top来说使用频率未必很高,甚至还可能略低于vmstat,top进入交互式界面后,按1显示所有cpu的负载,结果和mpstat基本一致,本文对mpstat使用做下简单的记录,方便回顾输出参数的含义。

mpstat使用介绍和输出参数详解

更新历史

2019年06月29日 - 初稿

阅读原文 - https://wsgzao.github.io/post/mpstat/

扩展阅读


mpstat简介

mpstat 是 Multiprocessor Statistics 的缩写,是实时系统监控工具。其报告与 CPU 的一些统计信息,这些信息存放在 /proc/stat 文件中。在多 CPU 系统里,其不但能查看所有 CPU 的平均状况信息,而且能够查看特定 CPU 的信息。

mpstat语法

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
# mpstat属于sysstat
yum provides mpstat
Loaded plugins: fastestmirror
Loading mirror speeds from cached hostfile
* base: mirror.vodien.com
* epel: download.nus.edu.sg
* extras: mirror.vodien.com
* updates: mirror.vodien.com
sysstat-10.1.5-17.el7.x86_64 : Collection of performance monitoring tools for Linux
Repo : base
Matched from:
Filename : /usr/bin/mpstat

sysstat-10.1.5-17.el7.x86_64 : Collection of performance monitoring tools for Linux
Repo : @base
Matched from:
Filename : /bin/mpstat


mpstat [-P {cpu|ALL}] [internal [count]]

其中,各参数含义如下:

参数 含义
-P {cpu l ALL} 表示监控哪个 CPU, cpu 在 [0,cpu 个数 - 1] 中取值
internal 相邻的两次采样的间隔时间
count 采样的次数,count 只能和 delay 一起使用

使用 mpstat 命令

1. 直接使用 mpstat 命令:

1
2
3
4
5
6
mpstat

Linux 3.10.0-862.el7.x86_64 07/04/2019 _x86_64_ (32 CPU)

11:45:19 AM CPU %usr %nice %sys %iowait %irq %soft %steal %guest %gnice %idle
11:45:19 AM all 0.39 0.00 0.19 0.00 0.00 0.03 0.00 0.00 0.00 99.39

当 mpstat 不带参数时,输出为从系统启动以来的平均值。

2. 使用 mpstat -P ALL 5 2 命令

1
2
3
4
5
6
7
8
mpstat -P ALL 5 2

Linux 3.10.0-862.el7.x86_64 07/04/2019 _x86_64_ (32 CPU)

11:44:11 AM CPU %usr %nice %sys %iowait %irq %soft %steal %guest %gnice %idle
11:44:16 AM all 0.38 0.00 0.21 0.00 0.00 0.03 0.00 0.00 0.00 99.38
11:44:16 AM 0 0.20 0.00 0.20 0.00 0.00 0.00 0.00 0.00 0.00 99.60
11:44:16 AM 1 0.20 0.00 0.20 0.00 0.00 0.00 0.00 0.00 0.00 99.60

表示每 5 秒产生一个报告,总共产生 2 个。

上图表示每 5 秒产生了 2 个关于处理器的统计数据报告,一共产生 2 个 interval 的信息,然后再给出这 2 个 interval 的平均信息。默认时,输出是按照 CPU 号排序。第一个行给出了从系统引导以来的所有活跃数据。接下来每行对应一个处理器的活跃状态。

输出参数含义

当没有参数时,mpstat 则显示系统启动以后所有信息的平均值。有 interval 时,第一行的信息自系统启动以来的平均信息。从第二行开始,输出为前一个 interval 时间段的平均信息。

输出各参数含义:

参数 释义 从 /proc/stat 获得数据
CPU 处理器 ID
%usr 在 internal 时间段里,用户态的 CPU 时间(%),不包含 nice 值为负进程 usr/total*100
%nice 在 internal 时间段里,nice 值为负进程的 CPU 时间(%) nice/total*100
%sys 在 internal 时间段里,核心时间(%) system/total*100
%iowait 在 internal 时间段里,硬盘 IO 等待时间(%) iowait/total*100
%irq 在 internal 时间段里,硬中断时间(%) irq/total*100
%soft 在 internal 时间段里,软中断时间(%) softirq/total*100
%steal 显示虚拟机管理器在服务另一个虚拟处理器时虚拟 CPU 处在非自愿等待下花费时间的百分比 steal/total*100
%guest 显示运行虚拟处理器时 CPU 花费时间的百分比 guest/total*100
%gnice gnice/total*100
%idle 在 internal 时间段里,CPU 除去等待磁盘 IO 操作外的因为任何原因而空闲的时间闲置时间(%) idle/total*100

CPU 总的工作时间:

1
2
3
4
5
6
7
total_cur = user + system + nice + idle + iowait + irq + softirq

total_pre = pre_user + pre_system + pre_nice + pre_idle + pre_iowait + pre_irq + pre_softirq

user = user_cur – user_pre

total = total_cur - total_pre

其中_cur 表示当前值,_pre 表示 interval 时间前的值。上表中的所有值可取到两位小数点。

Note:

  1. vmstat 和 mpstat 命令的差别:mpstat 可以显示每个处理器的统计,而 vmstat 显示所有处理器的统计。因此,编写糟糕的应用程序(不使用多线程体系结构)可能会运行在一个多处理器机器上,而不使用所有处理器。从而导致一个 CPU 过载,而其他 CPU 却很空闲。通过 mpstat 可以轻松诊断这些类型的问题。

  2. vmstat 中所有关于 CPU 的总结都适合 mpstat。当您看到较低的 % idle 数字时,您知道出现了 CPU 不足的问题。当您看到较高的 % iowait 数字时,您知道在当前负载下 I/O 子系统出现了某些问题。

文章目录
  1. 1. 前言
  2. 2. 更新历史
  3. 3. mpstat简介
  4. 4. mpstat语法
  5. 5. 使用 mpstat 命令
  6. 6. 输出参数含义